Handling missing values in multiple factor analysis

• We propose a new method to handle missing values in multiple factor analysis.
• The method handles missing values in continuous and categorical multi-table datasets.
• The methodology is available in the free software R.
• A sorting task where judges evaluate a subset of products illustrates the method.
• The method can manage rows of missing values in the multiple tables dataset.

Handling missing values is an unavoidable problem in the practice of statistics. We focus on multiple factor analysis in the sense of Escofier and Pagès (2008), a principal component method that simultaneously takes into account several multivariate datasets composed of continuous and/or categorical variables. The suggested strategy to deal with missing values, named regularised iterative MFA, is derived from a method available in principal component analysis which consists in alternating a step of estimation of the axes and components and a step of estimation of the missing values. The pattern of missing values considered can be structured with missing rows in some datasets. Some simulations and real examples that cover several situations in sensory analysis are used to illustrate the methodology. We focus on the important issue of the maximum number of products that can be assessed during an evaluation task.